摘要
手势识别一直是人机交互领域的一个重要研究方向.本文开展了基于微机电系统惯性测量单元实现手势灵活控制无人机技术的研究,设计开发了手势采集装置.提出了一种基于加速度计和陀螺仪传感器数据的深度学习框架ConvBLSTM:利用四个卷积层提取原始传感器数据的局部特征,为学习动态连续手势序列的时间特性,将卷积层提取的特征再输入双向循环层来获取全局时序特征,完成了高精度的手势分类.应用到工程项目中,定义了十种无人机手势指令来控制其飞行状态.所提出的方法平均达到了99.2%的高识别精度,相比经典的支持向量机、K近邻、长短时记忆网络等模式识别方法,解决了经典模型精度低、泛化能力差等问题.本文进行了多次实验来验证算法的适用性以及手势分类的准确性,通过数据点长度准确性实验,缩短了手势识别的滑动时间窗长度,提升了无人机实时响应速度.
Gesture recognition has always been a main branch in the research field of human-computer interaction.In this paper,we introduce a micro-electro-mechanical system(MEMS)-based gesture recognition approach and apply it on unmanned aerial vehicle(UAV)control.Six inertial measurement units are used to capture gestures,each of which is composed of a three-axis accelerometer and a three-axis gyroscope.We further propose ConvBLSTM framework,a deep learning framework based on accelerometer and gyroscope sensor data via four convolution layers to extract the local features of the raw sensor data.To learn the temporal characteristics of the continuous gesture sequence,the features extracted from convolution layer are input the Bi-directional Long Short-Term Memory network layer to obtain global temporal features.Furthermore,we define ten types of gesture commands to control the UAV flight state,including rise,fall,forward,backward,left,right,left,right,hover and non-control states.Compared to the SVM,KNN,BiLSTM and other classical pattern recognition methods,the proposed method solves the problems of low accuracy and poor generalization ability of the classical model.The proposed gesture recognition classifier successfully distinguished ten gestures,achieving a high recognition accuracy of 99.2%.In this paper,multiple experiments were conducted to verify the applicability of the algorithm and the accuracy of gesture classification.Shorten the sliding window length of gesture recognition through data point length accuracy experiments,Improve real-time response speed of UAV.
作者
刘璇恒
邓宝松
裴育
范博辉
谢良
闫野
印二威
LIU Xuan-heng;DENG Bao-song;PEI Yu;FAN Bo-hui;XIE Liang;YAN Ye;YIN Er-wei(School of Software,Tianjin University,Tianjin 300350,China;Unmanned Systems Research Center,National Institute of Defense Technology Innovation,Academy of Military Sciences China,Beijing 100071,China;School of Software,Beihang University,Beijing 100191,China;School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China;Tianjin Artificial Intelligence Innovation Center(TAIIC),Tianjin 300450,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2020年第11期2241-2248,共8页
Journal of Chinese Computer Systems
基金
国家自然科学基金青年项目(61703407)资助
科技委智创基金项目(18-163-00-TS-003-007-01)资助.
关键词
手势识别
人机交互
微机电系统
深度学习
无人机
gesture recognition
human-computer interaction
micro-electro-mechanical system(MEMS)
deep learning
unmanned aerial vehicle(UAV)
